Last week, I wrote about Oliver Taylor, a bogus journalist generated with the help of a deep learning algorithm. (Here: https://www.reuters.com/article/cyber-deepfake-activist/deepfake-used-to-attack-activist-couple-shows-new-disinformation-frontier-idUSL1N2CX06O)

Reuters has a guide out on recognizing this kind of imagery; but I wanted to complement it with examples from http://thispersondoesnotexist.com 

So how do you spot static deepfake imagery in the wild?

Start by looking at the background. Generative adversarial networks still struggle to come up with convincing backdrops & they can get quite surreal:

In particularly horrifying cases, you can find misshapen bits of *other people's faces* lurking in the background. This is an unusually obvious example:

If you're looking at a female face, check the earrings. Oftentimes they'll only sport one, or the earrings won't match.

Look at the clothes. GANs still struggle with these. Bizarre collars, crazy hats, clothes that mix and merge together, or seem to be composed of different fabrics:

Check for unsightly ears! GANs sometimes make ears that look Elfin (left) or batlike (right).

Examine the hair. Note how the curly locks on the GAN-created image to the left seem to sprout out of nowhere or grow into inorganic-looking matrices. Or how the hair to the right appears to dread and then turn into ... a hat?

Put yourself in the mind of your dentist and have a good look at those gnashers because GANs struggle with teeth. Look closely at these and see how they seem to meld, blur, and coagulate. (Also, what on earth is that guy wearing?)

As with earrings, GANs have problems keeping eyeglasses symmetrical. So it's not unusual to see people wearing glasses with clashing frames, like this one:

THE BLOB.

For whatever reasons StyleGAN in particular occasionally deposits a tell-tale blob somewhere near the face. Can you spot the teeny water droplet-like glitch in the images below?

There're plenty of other signs to look out for but this'll have to do for now. As @HaoLi81 told me earlier this year, the technology is fast improving. These glitches may look obvious now, but who knows whether they'll still be around in a year's time. https://graphics.reuters.com/CYBER-DEEPFAKE/ACTIVIST/nmovajgnxpa/index.html